Classification of rice processing precision is an important link in rice processing. In order to accurately identify the grade of rice processing precision, by combining the hyper column technology (HCT), max-relevance and min-redundancy (MRMR) feature selection algorithm and extreme learning machine (ELM) technique, an improved VGG16 convolutional neural network was proposed. First of all, the OneHot format in machine learning was used for coding and normalization of data;then, combining HCT, the VGG16 convolutional neural network was used as the feature extractor, which can extract local differentiating features from deep structure at different levels. Totally 5248 rice features were extracted, the MRMR feature selection algorithm was employed to eliminate massive redundant rice image features, and 500 most effective features were selected. Finally, the ELM technique was used to classify the processing grade of rice. The 5848 sample images were randomly divided into the training set, test set and verification set according to the ratio of 6∶3∶1 for training and test of model. The results showed that when the rice processing grade classification model built based on the improved VGG16 convolutional neural network was used to classify the 1755 rice samples in the test set, the overall accuracy can reach 97.32%, and the classification prediction speed of rice processing precision can reach approximately 85t/h, which basically satisfied the requirement of rice production line.